You are on page 1of 8

Downloaded from rsta.royalsocietypublishing.

org on June 2, 2011

Phil. Trans. R. Soc. A (2011) 369, 2414–2421


doi:10.1098/rsta.2011.0090

Particle stress in suspensions of soft objects


BY T. KRÜGER1, *, F. VARNIK2 AND D. RAABE1
1 Max-Planck-Institut für Eisenforschung, Max-Planck-Strasse 1,
40237 Düsseldorf, Germany
2 Interdisciplinary Centre for Advanced Materials Simulation, Stiepeler Strasse
129, 44780 Bochum, Germany

In a suspension of extended objects such as colloidal particles, capsules or vesicles, the


contribution of particles to the stress is usually evaluated by first determining the stress
originating from a single particle (e.g. via integrating the fluid stress over the surface of a
particle) and then adding up the contributions of individual particles. While adequate for
a computation of the average stress over the entire system, this approach fails to correctly
reproduce the local stress. In this work, we propose and validate a variant of the method
of planes which overcomes this problem. The method is particularly suited for many-body
interactions arising from, for example, shear and bending rigidity of red blood cells.
Keywords: lattice Boltzmann method; immersed boundary method; haemorheology;
apparent viscosity; particle stress; method of planes

1. Introduction

Understanding the rheology of dense suspensions and emulsions is a complex field


in modern physics and engineering. For example, it is of practical importance
to find reliable constitutive laws for the rheology of suspensions of deformable
particles, e.g. capsules, vesicles and red blood cells (RBCs), derived from
microscopic properties such as particle shape and deformability. One quantity
of central interest in such investigations is the stress tensor. The total stress is
the sum of the fluid and particle stresses, s(r, t) = sf (r, t) + sp (r, t). Usually,
the evaluation of the local fluid stress sf (r, t) does not pose a serious problem
[1–3]. The major difficulty is the computation of the local particle stress sp (r, t).
One may follow Batchelor [4] and use the fluid stress information at the
surface of the particles or—equivalently—employ the approach presented by
Ramanujan & Pozrikidis [5], who use the known membrane forces instead. When
summed over all the particles, the correct volume-averaged particle stress may
then be obtained. However, as will be shown in this paper, such ‘particle-based’
methods yield incorrect results on the local stress.
A common way to circumvent this problem is to study simple cases in which the
macroscopic stress is constant over the entire channel cross section (e.g. planar
Couette flow) or is known from analytical solutions of the momentum balance
*Author for correspondence (t.krueger@mpie.de).

One contribution of 25 to a Theme Issue ‘Discrete simulation of fluid dynamics: applications’.

2414 This journal is © 2011 The Royal Society


Downloaded from rsta.royalsocietypublishing.org on June 2, 2011

Particle stress in suspensions 2415

equation, as in the case of a Poiseuille flow (e.g. [6]) or Kolmogorov flow (e.g. [7]).
A drawback of such a detour is that fluctuations of local shear stress are not
accessible, thus setting severe limits to an analysis of the underlying problem
within the framework of statistical physics (recall, for example, that equilibrium
fluctuations of shear stress yield an independent measure of shear viscosity).
The present article focuses on this issue for the case of suspensions of soft
objects confined between two planar walls. Here, the term ‘soft’ means that
interaction potentials are differentiable so that a force can always be defined.
We propose a method (a variant of the method of planes (MOP)) that allows
the particle contribution to the local stress at any desired distance from a solid
wall to be directly computed from interaction forces. As will be shown below, this
method accurately reproduces the expected macroscopic behaviour of the local
viscous shear stress while at the same time providing access to spatio-temporal
fluctuations of this important quantity.

2. Stress evaluation

We study a dense suspension of RBCs immersed in a Newtonian fluid with shear


viscosity h0 , which fills both the interior and exterior regions of the particles.
Our simulation algorithm consists of a fluid solver, a membrane model and a
fluid–membrane coupling scheme. For the fluid, we use the D3Q19 Bhatnagar–
Gross–Krook lattice Boltzmann method (LBM) [8] with a linearized equilibrium
function in order to avoid inertia [9]. The interaction of the fluid and the mem-
brane is captured by the immersed boundary method (IBM) [10]. The membrane
dynamics is derived from a constitutive model, and the strains in the membranes
are evaluated using a finite-element method. A detailed description of the
simulation method can be found in Krüger et al. [9].
As already mentioned, in a particulate suspension, the total stress is the sum of
the fluid and particle stresses. Within the LBM, it is straightforward to evaluate
the fluid stress locally (e.g. [2,3]). We thus focus here on a computation of the
particle stress at a given position and independent of any further assumptions.
The effect of interactions on flow behaviour can be incorporated in the Navier–
Stokes equations either (i) by direct implementation of particle forces as a
(spatially and temporally varying) external force field or (ii) by introducing the
particle stress tensor. The relation between membrane force density acting on the
fluid and particle stress is
f (r, t) = V · sp (r, t). (2.1)
In the following, we restrict the discussion to the case of a suspension confined
between two parallel walls, separated by Lz along the z-direction. Noting that

f (r, t) = Ni F i (t)d(r − r i (t)) (F i is the force on particle i and r i is its position)
and integrating equation (2.1) along the x- and y-directions, one arrives at

1
N
v p
Fia (t)d(z − zi (t)) = Saz (z, t), a = x, y, z, (2.2)
A i vz
L L
where we defined Saz p p
(z, t) := 0 x 0 y dx dysaz (r, t)/A, where A = Lx Ly (Lx and
Ly being the lengths of the system along the x- and y-directions). Assuming that

Phil. Trans. R. Soc. A (2011)


Downloaded from rsta.royalsocietypublishing.org on June 2, 2011

2416 T. Krüger et al.

(a) (b)
wall z z

z-axis
p
Figure 1. Sketch of the method of planes. The particle stress Saz (z, t) is computed. (a) (Lagrangian
frame): membrane nodes that are on the right side of the plane (zi > z) contribute with their
negative force; nodes on the left side (zi < z) with their positive force; see equation (2.3).
(b) (Eulerian frame): lattice nodes that are on the right side of the plane (Zj > z) contribute with
their negative force; nodes on the left side (Zj < z) with their positive force; see equation (2.4).
(Online version in colour.)

the particle contribution to the stress vanishes at the wall (which is always the
case here), equation (2.2) can be integrated with respect to z. This gives
N 
1 z  1 
N

Saz (z, t) =
p
dz Fia (t)d(z − zi (t)) = Fia (t) sgn(z − zi (t)), (2.3)
A i 0 2A i

where sgn is the sign function. In deriving equation (2.3), we used the fact
that the volume integral over particle force is identically zero (momentum
conservation). Equation (2.3) gives the stress on a plane at z parallel to the walls
(figure 1).
The above approach is inspired by the so-called ‘MOP’, first introduced by
Todd et al. [11] for the case of a simple liquid and further examined by Varnik
et al. [12] in the case of a polymer melt. To the best of our knowledge, an
extension of the MOP to the case of extended particles immersed in a fluid has
not been proposed so far. The key point motivating the use of equation (2.3)
in the case of interest to us is to realize that each individual membrane object
(capsule, RBC) is represented by an ensemble of interacting nodes immersed in
the ambient fluid. If we identify these membrane nodes as point particles, the
fact that the thus defined point particles belong to extended objects plays no
role in a computation of the particle stress. Thus, the index i in equation (2.3)
specifies a node, and the sum runs over all the membrane nodes in the system.
Note that the forces F i in equation (2.3) do not necessarily have to be pair
forces. This is essential for the applicability of the method in the present case
since strain and bending resistances of the membranes effectively originate from
many-particle interactions.
A priori, the fluid stress is known on the Eulerian lattice while forces needed for
the evaluation of the particle stress are computed on the Lagrangian mesh. Owing
to the fact that—in the present study—both systems are coupled by non-local
IBM interpolations [9,10], this results in a weak but observable numerical noise of
the total stress (data not shown). This problem is solved by first computing, via
IBM interpolations, particle forces on the Eulerian grid and then applying the

Phil. Trans. R. Soc. A (2011)


Downloaded from rsta.royalsocietypublishing.org on June 2, 2011

Particle stress in suspensions 2417

p
MOP to obtain the particle stress Sxz (z, t) completely in the Eulerian system.
Equation (2.3) then reads
1 
p
Sxz (z, t) = Fjx (t) sgn(z − Zj ), (2.4)
2A j

where the sum runs over all lattice nodes j and Fjx (t) is the x-component of the
force acting on node j at time t. This approach is reasonable since membrane
forces are ‘visible’ through their action on the fluid only. As the fluid stress
is computed at the fluid nodes, the particle stress should also be evaluated at
those coordinates. For this purpose, we employ the relation Sxz p
(Z , t) = (Sxz
p
(Z +
DZ /2, t) + Sxz (Z − DZ /2, t))/2, i.e. we take the average of two planes separated
p

by one lattice constant DZ , where each plane is in the middle of two lattice grid
lines (see figure 1). It is easy to see that this approach is equivalent to applying the
MOP directly to a plane located at position Z on the Eulerian grid, ignoring all
nodes on that plane. This is because, within the former approach, the contribution
of the nodes directly on the plane at Z is taken twice but with different signs,
thus leading to its cancellation.

3. Results

We consider the flow between two planar walls placed along the z-axis at positions
z = 0 and z = D and moving along the x-axis with velocities ±uw , respectively (no
external forces). This defines a global shear rate of ġ = 2uw /D. In such a set-up,
it follows from the momentum balance equation that the xz-component (shear
component) of the macroscopic deviatoric stress tensor is constant along the
z-direction, sxz (z) = const. (‘macroscopic’ means that instantaneous fluctuations
are averaged out). It is important to realize that this relation is valid for any flow
profile and constitutive law of the fluid. It thus provides an excellent test case for
our approach. In the following, all simulation quantities are given in lattice units,
i.e. DZ = 1, Dt = 1 and r = 1.
We have performed six series of simulations with wall-to-wall separations
D = 15, . . . , 90 in steps of 15. The size of the simulation boxes, including the
bounce-back walls, is 90 × 90 × (D + 2) (table 1), with full periodicity along the
x- and y-axes. The shear flow is imposed by moving bottom and top walls at z = 0
and z = D via the bounce-back boundary condition [1]. In order to maintain the
same average shear rate, the wall velocity increases linearly with the wall-to-
wall distance and is 0.06 for the largest simulation box (table 1). Each series
consists of 10 independent simulation runs with different random initial RBC
configurations. The results in each series have been averaged in order to improve
the ensemble averages. The LBM relaxation parameter is t = 1, and the two-point
IBM interpolation stencil has been used [9]. The RBCs have a large radius r = 9
and are discretized using 1620 triangular faces each (figure 2). The number of
RBCs has been chosen in such a way that the haematocrit is 46.7 per cent in
all studied cases (table 1). Each simulation ran for 50 000 time steps (according
to nearly 67 inverse shear rates). The first 10 000 time steps have been omitted,
thus neglecting transients. The remaining data have been used for time averaging
(observables have been evaluated at every 100th time step).

Phil. Trans. R. Soc. A (2011)


Downloaded from rsta.royalsocietypublishing.org on June 2, 2011

2418 T. Krüger et al.

(a) (b)

Figure 2. (a) Membrane mesh of a RBC with 1620 triangular faces. The large (small) diameter of
a RBC is 18 (approx. 6) lattice units. (b) Snapshot of a simulation with channel width D = 90 and
300 RBCs. (Online version in colour.)

Table 1. Simulation parameters and results. Collection of the lattice size, the number of RBCs, the
p
wall velocity uw , the volume- and time-averaged particle shear stress sxz , and the resulting relative
apparent viscosity h/h0 (h0 is the viscosity of the pure ambient fluid, i.e. in the absence of RBCs)
as a function of wall distance D. The volume-averaged fluid stress is 2.2̄ × 10−4 in all cases studied.
Note that the relative apparent viscosity is nearly identical for all wall-to-wall separations.

p
D lattice size RBCs uw sxz h/h0

15 90 × 90 × 17 50 0.01 9.42 × 10−4 5.2


30 90 × 90 × 32 100 0.02 9.83 × 10−4 5.4
45 90 × 90 × 47 150 0.03 1.00 × 10−3 5.5
60 90 × 90 × 62 200 0.04 1.01 × 10−3 5.6
75 90 × 90 × 77 250 0.05 1.01 × 10−3 5.6
90 90 × 90 × 92 300 0.06 1.01 × 10−3 5.6

Figure 3 contains the central result of our work. It compares the present
method with a computation of local shear stress using ‘particle-based’ approaches,
i.e. by computing the stress over each individual cell and then adding the
contributions of all the cells whose centre of mass is within the interval [Z −
1/2, Z + 1/2]. While the latter approach fails to satisfy the requirement of a
constant macroscopic shear stress across the channel, the proposed method
perfectly fulfils this condition. We have seen this behaviour in all the cases
studied. Generally, the ‘particle-based’ total stress reflects oscillations of the
particle density (data not shown), while the new approach is free of this artefact.
Indeed, such an ad hoc definition of the local stress must reflect, at least to some
extent, the concentration profile of particles [12].
We emphasize that, when averaged over the entire simulation cell, both
methods yield the same result—up to relative deviations of the order of 10−3
or less. The basic difference is thus in the evaluation of the local shear stress.
This, however, is a major point when it comes to determining the local shear
viscosity, which is the ratio of the local shear stress and the local rate of shear.
As mentioned in §1, the knowledge of macroscopic stress may help to alleviate

Phil. Trans. R. Soc. A (2011)


Downloaded from rsta.royalsocietypublishing.org on June 2, 2011

Particle stress in suspensions 2419

(a) (b)
z=0
0.0025 0.0025 z = 45.5
wall stress 0.0014

total stress
z = 90
total stress 0.0013
0.0020 fluid stress 0.0020
particle stress 0.0012
stress (LU)

0.0015 10 000 20 000 30 000 40 000 50 000


0.0015 time (LU)

0.0010 0.0010

0.0005 0.0005

0 15 30 45 60 75 90 0 15 30 45 60 75 90
z (LU) z (LU)

Figure 3. Stress profiles obtained within (a) the ‘particle-based’ method and (b) the present
approach for the case of a dense suspension of RBCs (46.7% volume fraction) in a planar Couette
cell (wall-to-wall distances D = 90 ≈ 15d, d denoting the small diameter of a RBC). The time-
averaged particle, fluid and total stresses are shown together with the wall stress. The solid lines
are intended as a guide. The legend in (a) is also valid for (b). Note that, in the present flow
situation, the total stress—when averaged over fluctuations—is expected to be constant across the
channel. The inset in (b) shows the temporal fluctuations of the total stress at different z-positions
as indicated. Note that the kinetic contribution to the stress is disregarded both in (a) and (b)
since the Reynolds number is negligible. (Online version in colour.)

50
D = 15
relative apparent viscosity

40 D = 30
D = 45
D = 60
30 D = 75
D = 90
20

10
increasing D
0 5 10 15 20 25 30 35 40 45
z (LU)

Figure 4. Relative apparent viscosity profiles, h(z)/h0 (h0 is the fluid viscosity in the absence of
RBCs), for all tested wall-to-wall distances D as a function of transverse coordinate z. Since the
flow problem is symmetric, the profiles have been symmetrized and are only shown for one half of
the channel (between 0 and D/2). The solid lines are intended as a guide. As the channel width
decreases, the amplitude of wall effects increases and the position of the maximum viscosity is
slightly shifted towards the wall. (Online version in colour.)

this problem. It does, however, not provide any access to the fluctuations of the
local stress. In contrast to this, fluctuations of the local stress are a natural result
of the present method (see the inset of figure 3b).
In figure 4, wall effects can easily be identified. In particular, the local viscosity
is highest in a region close to the wall. The maximum and the location of this
viscosity peak depend on the wall distance D. For a similar model to that in

Phil. Trans. R. Soc. A (2011)


Downloaded from rsta.royalsocietypublishing.org on June 2, 2011

2420 T. Krüger et al.

(a) (b)
0.0012 wall stress
total stress
fluid stress
particle stress
0.0008

stress (LU)
0.0004

0 5 10 15 20 25 30
z (LU)

Figure 5. Simulation of a single capsule in shear flow (radius of the undeformed capsule is r = 9,
the channel width (wall-to-wall distance) is D = 30, the velocity of either wall is uw = ±0.05).
(a) Snapshot of the simulation and (b) stress distribution. The solid lines are intended as a guide.
The kinetic contribution to the stress is disregarded since the Reynolds number is negligible. (Online
version in colour.)

the present work, Doddi & Bagchi [6] also observed an increase in viscosity
near the walls in a Poiseuille flow. Varnik & Binder [13], on the other hand,
reported a significant increase (by two orders of magnitude) in the local viscosity
of a polymer melt close to a solid substrate. However, suspensions of deformable
particles usually exhibit a shear-thinning behaviour (e.g. the experiments of Chien
[14] on RBCs), which depends on the applied shear rate or stress. In a Poiseuille
flow, the stress varies linearly with distance from the wall and thus leads to a
locally variable shear thinning [15], which superimposes the wall effects. Indeed,
since the stress increases when approaching the walls, this would contribute to a
corresponding decrease in viscosity owing to shear thinning. The observed increase
in local shear viscosity in fig. 15 of Doddi & Bagchi [6] thus results from the
competition between two opposite effects: the confinement effects—leading to an
increase in local viscosity when approaching the wall—as opposed to variable
stress effects that tend to decrease it. In the present study, on the other hand,
the macroscopic stress is constant across the channel, which allows a separation
of wall effects from influences related to shear thinning.
The strength of the present approach is further highlighted in figure 5. Clearly,
the proposed method reproduces the expected constant macroscopic shear stress
profile even in the case of a single capsule. Note that, in this case, the methods
introduced by Batchelor [4], Ramanujan & Pozrikidis [5] and Ladd [1] would yield
a single datum point as the average over the entire capsule.

4. Conclusions

A new approach for the evaluation of the local particle stress in immersed
boundary lattice Boltzmann simulations is proposed and tested. It is possible
to compute the particle stress as a function of the transverse coordinate
independently of macroscopic assumptions. The spatial resolution is the same
as that for the viscous fluid stress in the LBM.

Phil. Trans. R. Soc. A (2011)


Downloaded from rsta.royalsocietypublishing.org on June 2, 2011

Particle stress in suspensions 2421

The proposed method accurately recovers the constancy of shear stress in


simple shear flow, even for a dense suspension of RBCs (46.7% volume fraction)
with highly inhomogeneous viscosity. The volume average of the shear stress
obtained with the newly proposed method equals the values calculated from two
other methods which, however, fail to provide the correct behaviour of the local
stress.
Even though there are ways, in the case of simple flow situations, to obtain
the macroscopic average of the local shear stress, these approaches do not
provide any access to stress fluctuations—a quantity of great importance within
the framework of statistical physics. The problem is also solved within the
present approach.
Finally, we show that the method works perfectly even in the case of a single
object in a linear shear flow.
This project has been supported by DFG grant VA205/5-1.

References
1 Ladd, A. J. C. 1994 Numerical simulations of particulate suspensions via a discretized
Boltzmann equation. I. Theoretical foundation. J. Fluid Mech. 271, 285–309. (doi:10.1017/
S0022112094001771)
2 Krüger, T., Varnik, F. & Raabe, D. 2009 Shear stress in lattice Boltzmann simulations. Phys.
Rev. E 79, 046704. (doi:10.1103/PhysRevE.79.046704)
3 Krüger, T., Varnik, F. & Raabe, D. 2010 Second-order convergence of the deviatoric stress
tensor in the standard Bhatnagar–Gross–Krook lattice Boltzmann method. Phys. Rev. E 82,
025701(R). (doi:10.1103/PhysRevE.82.025701)
4 Batchelor, G. K. 1970 The stress system in a suspension of force-free particles. J. Fluid Mech.
41, 545–570. (doi:10.1017/S0022112070000745)
5 Ramanujan, S. & Pozrikidis, C. 1998 Deformation of liquid capsules enclosed by elastic
membranes in simple shear flow: large deformations and the effect of fluid viscosities. J. Fluid
Mech. 361, 117–143. (doi:10.1017/S0022112098008714)
6 Doddi, S. K. & Bagchi, P. 2009 Three-dimensional computational modeling of multiple
deformable cells flowing in microvessels. Phys. Rev. E 79, 046318. (doi:10.1103/PhysRevE.
79.046318)
7 Janoschek, F., Mancini, F., Harting, J. & Toschi, F. 2011 Rotational behaviour of red blood
cells in suspension: a mesoscale simulation study. Phil. Trans. R. Soc. A 369, 2337–2344.
(doi:10.1098/rsta.2011.0086)
8 Qian, Y.-H., d’Humieres, D. & Lallemand, P. 1992 Lattice BGK models for Navier–Stokes
equation. Europhys. Lett. 17, 479–484. (doi:10.1209/0295-5075/17/6/001)
9 Krüger, T., Varnik, F. & Raabe, D. 2010 Efficient and accurate simulations of deformable
particles immersed in a fluid using a combined immersed boundary lattice Boltzmann finite
element method. Comput. Math. Appl. (doi:10.1016/j.camwa.2010.03.057)
10 Peskin, C. S. 2002 The immersed boundary method. Acta Numer. 11, 479–517. (doi:10.1017/
S0962492902000077)
11 Todd, B. D., Evans, D. J. & Daivis, P. J. 1995 Pressure tensor for inhomogeneous fluids. Phys.
Rev. E 52, 1627–1638. (doi:10.1103/PhysRevE.52.1627)
12 Varnik, F., Baschnagel, J. & Binder, K. 2000 Molecular dynamics results on the pressure tensor
of polymer films. J. Chem. Phys. 113, 4444–4453. (doi:10.1063/1.1288390)
13 Varnik, F. & Binder, K. 2002 Shear viscosity of a supercooled polymer melt via non-equilibrium
molecular dynamics simulations. J. Chem. Phys. 117, 6336–6349. (doi:10.1063/1.1503770)
14 Chien, S. 1970 Shear dependence of effective cell volume as a determinant of blood viscosity.
Science 168, 977–979. (doi:10.1126/science.168.3934.977)
15 Varnik, F. & Raabe, D. 2008 Profile blunting and flow blockage in a yield-stress fluid: a
molecular dynamics study. Phys. Rev. E 77, 011504. (doi:10.1103/PhysRevE.77.011504)

Phil. Trans. R. Soc. A (2011)

You might also like